I’ve gotten into an interesting discussion on a mailing list at work – the debate is in part about whether we should care if a search engine results page takes an extra half second to load and render, vs. rendering almost instantaneously.
(Now, bear in the mind that this mailing list is so random in composition and topic that it even has “random” in its name. It’s a random assortment of Y! engineers venting randomly about whatever random thing has gotten up their collective or individual noses lately. And most important of all, these are mostly not people who study user behavior full-time, and (mostly) not people who are going to affect our own render time for search result pages…. but the reactions are interesting.)
Several people have opined that load/render time of 0.5 seconds vs., say, 0.05 seconds couldn’t possibly make any difference to users. What impresses me most about this reaction is not the opinion itself, but the certitude — they seem to be 100% sure that differences of this magnitude can’t matter, based on some combination (I assume) of introspection and back-of-the-envelope thinking.
Myself, I believe that the shorter the timescales get, the less helpful introspection is going to be, and the more we’re in the purview of experimental psychology. I mean, I know that I become annoyed during the many-second wait for Technorati search results to display, and I think my self-report of annoyance should carry some weight. But I also know that I don’t know exactly how I return a ping-pong serve or block a foosball shot, and that an expert could well tell me something I don’t know about what’s going on during those tenths of a second.
I have seen a study indicating that a 50-millisecond (i.e. 0.05 second) exposure to a web page is sufficient for a user to form a visual like/dislike opinion. Now, take a time that is ten times that long (0.5 seconds), and ask yourself – is it impossible that users form a like/dislike opinion during that interval based on the failure of the page to finish rendering?
My own guess (and here I’m just speculating wildly) is that on any one search it doesn’t make any difference whether the page renders in 50 msecs or 500 msecs. But over time, particularly with repeated sequential queries, there could be a cumulative effect where the service feels slightly draggy at 500 msec (and refreshingly zippy at 50 msec) and that this could affect people’s affection for the product, _even_ if they cannot articulate what it is they like or don’t like. What do you think?
Tim Converse 
I agree with you, Tim. It makes sense to me that with a bunch of searches in a row, half a second here and there will feel slow.
I think half a second might matter. I’ve also seen the study which said that people judge web page quality in 0.05 seconds — my recollection of it is not that they like or dislike it in that short span of time, but they accurately predict what the overall opinion is about the quality of the web page in that time. That’s a much more interesting conclusion (and I hope I’m remembering it right). While it doesn’t necessarily mean that a half a second wait for a page to download and render dooms the site, it does indicate that people do notice.
(I can’t resist quoting Carrie Fisher here: “Immediate gratification takes too long.”)
I agree that over time even a half second would be noticeable assuming the “other search” was giving similar results in less time. However I’d be happy to wait longer if the results were better, especially for really specific and complex queries.